Yes, REW is my tool of choice too. The Realiser convolves with a DSP chip inside and as far as I can tell it uses proprietary formats. So you can measure with the provided microphones and then there are (quite restricted) options to truncate the IR and to merge with a flat one in bass (80Hz or 120Hz merging point). But that is it, unless there is something I did not find. If only one could open the "PRIR" as a wav modify and load back!
I searched for the manual on Google and skimmed it for a bit, but all the endless English terms suddenly gave me a headache. It really seems to use its own proprietary format. =(
There seem still to be a misunderstanding, I understand the need for head tracking, but you talked about the difference in "speaker head tracking" and "earphone head tracking". Or did you talk about speaker based cross talk cancelation like in Ambiophonics/Bacch?
It was both.
I’ll admit I see the value in head tracking, but when I think it through, I wonder: with headphones we’re deliberately adding dynamic motion and changes—aside from the slight boost to externalization, is that really necessary? A few years ago I asked another user, “If you move your body wildly, will head tracking correct all of that?” Their answer was that there are definite limits.
So I asked myself: what if, with speakers, head tracking were truly perfect and the imaging remained locked in place? Isn’t that essentially what a headphones/IEM BRIR does? No matter how much you shake your head or even stand on your hands, the sound stays fixed. That was just a passing thought—not a major point.
This is interesting. If I understand correctly you started with the green IR and tweaked to get to the pink one?
If so, how did you do it? Did you cut the IR into pieces and reassembled after changements, or is there a clever function to do this in one go?
Yes. That was probably someone else’s response—I grabbed the example from my folder and captured what I’d posted in the Korean community. And yes, I started with the green IR and tweaked it into the pink one.
Since this example is about cutting down reflections, I’ll keep this brief. In this case, my (or the other person’s) goal was to reduce reflections so much that it’s almost anechoic.
Completely removing reflections is easy—just gate the IR and you’re done. (Of course, that chops off the low end.)
Another approach is the MTW feature in the REW beta: it lets you apply different gating settings to each frequency band.
Something like this: MTW can be applied very simply. After all, you can correct or even synthesize the low end. Usually you only need about 5 ms—roughly up to 4 ms of the impulse response—so you can export it that way. The problem is that, in reality, even an anechoic chamber won’t decay that fast (since there are no reflections).
The reason I don’t worry much about the low-frequency band is that you can just synthesize it, and even though torso and chest reflections clearly contribute, you can still rely on the pattern of your original response for those as well.
In any case, you can do it easily like this, or experiment with other methods.
It’s like cancelling specific reflections or room modes with something like ART. But it’s surprisingly sensitive—low frequencies aren’t a big issue (and honestly it’s easier to just synthesize those), whereas high frequencies are much more finicky.
In an ideal room where side walls, front wall, back wall, floor, and ceiling are all spaced out at precisely the right time intervals and you can observe them in the impulse response, it’d be fairly easy to control. But in a typical home environment, that’s not the case—most reflections from all around hit you almost simultaneously.
So a simpler—but more realistic—approach than MTW is to literally attenuate the reflections’ response. How much you do this depends on the speaker’s directivity, the anechoic‑room conditions, and your goals, but in the example I shared, “realistic” meant leaving only the direct sound.
drive.google.com
If you follow the link, you’ll likely find a text file there.
To keep it simple, apply a short MTW to your original IR and export it, then re-import it. That way, all the reflections will be cut off.
And if you apply the TXT file you downloaded from the link as the calibration curve to the MTW‑processed IR, you’ll see something like that. That’s essentially the target.
And then, on the original response (with no MTW applied), you gate the reflection portion, apply a bit of smoothing, and then EQ it. In the example, I just used REW’s Auto EQ. Since the goal is to kill the reflections and push them below the audible threshold, there’s no need for precision—just make it look reasonably good.
And when you combine the EQ’d response with the MTW‑processed IR you align(or A+B), the reflections end up attenuated in a realistic, almost anechoic way.
That’s just the start—you can also tweak individual reflections or even shape the ETC to your liking.
There’s a lot you can do with REW, but since almost no one uses REW for BRIR, it’s mostly just me (and a few Korean users) running repeated experiments and validations against the theories in published acoustics papers.
There’s a lot you can do this way, but most people want a “one‑click solution.” In Korea, almost nobody actually rolls up their sleeves, experiments, and then tries it again—I know of only one or two who do. (I can’t speak for PhD‑level experts, but in the general audio community, they’re virtually nonexistent.) So I’ve spent countless hours studying BRIR, replicating paper methods, and validating them—thousands, even tens of thousands of times—and whenever I hit a snag, there was no one to ask and nowhere to find the answers. That’s why I started this thread: I’m certain there are others like me around the world.
The way I see it, that is not necessarily so. In the time domain one can modify the time slices more or less independently (in theory) and even during recording early reflections can be tweaked without change of power response (later reflections).
With the placement as far away as possible from all walls I arrived at this.
So far the sound is not bad at all for me with this BRIR (L_channel->L_ear)
What I meant was before any post-corretcion. I was referring to treating reflections in the real world through physical absorption.
Much more reflections than in the BRIRs you showed above. Interesting.
The BRIR for cross feed (R_channel->L_ear) looks similar and I experimented with attenuating crossfeed and as well with "cutting off" the initial peak (direct sound) completely with a suiting window to reduce the inherent stereo flaws a bit (creating ambiophonics instead of stereo). Still working on it.
It’s actually quite easy to emulate a simple physical barrier in a BRIR. Muting the opposite‑ear channel entirely might seem intuitive, but in practice it doesn’t work that way and is far too extreme. Here’s the key:
Imagine a 30° angle: the ILD will vary across frequency bands. Adjust your EQ to achieve a uniform attenuation of about –16 dB, but leave everything below roughly 700–800 Hz untouched. You can easily approximate this with a simple EQ. (To check it more easily, it helps to view the response in normalized mode.)
You can apply it just to the direct sound, or include reflections—but in my experience, when you stray from what’s realistically possible or necessary, you can feel that something’s off even in the BRIR. In other words, BRIR post‑processing offers tremendous freedom, but if you indulge in that freedom, you can end up doing things you shouldn’t—things that are impossible in reality.
For example: if no direct sound is played, how can any reflections exist? If you remove the direct‑sound peak in the opposite‑ear channel—i.e. no direct sound was ever played or recorded—how would any reflections be generated?
Of course it’s technically possible to post‑process, synthesize, and create it that way, but it ceases to reflect a realistic acoustic scenario.
I was listening to binaural recordings (most of the time made with Neumann KU100) with headphones and without my BRIR. The idea being to hear with "KU100's ears". The result was for instance for the basketball scene that everything happened in the back of my head. I remember a recording of a jazz band where the players all sounded to me as if they were placed hanging up at the ceiling in the corners. But at least they were kind of frontal.
It never occurred to me to use a BRIR on top of the recording through dummy ears. Seemed like putting ears on ears to me.
I listened to the recording you linked. It is not clear to me where the dummy head is placed, I cannot see it in the video. I only see ORTF stereo mics hanging above the orchestra. So it is not clear what to expect "realistically" in respect to clarity, proximity and spatial differentiation.
The sound is quite nice, more like in a seat (no surprise
than what a typical stereo recording tries to achieve. No surprise because a binaural recording from a good seat would have too much reverb for stereo speaker reproduction.
Everything is a compromise, a balance and a "product". My preferred place for a binaural recording would probably not be an actual seat either.
Interestingly there was an improvement when routing it through my BRIR (direct, no additional crossfeed).
Without BRIR the orchestra was rather fuzzy and the saxophone was floating somehow above the orchestra.
With the BRIR active the orchestra got more "body" and everything became clearer (similar to using a personal BRIR in room simulation) but the sax still was a few meters above. How realistic the imaging of the orchestra is, I cannot tell as the reference is not there.
Yes. Depth and spatial elements are what characterize binaural sources or recordings. Of course, you can listen with headphones or IEMs with nothing applied. But as I mentioned earlier, IEMs bypass the Pinna and headphones sit too close, so the sound tends to hover around your head and internalize easily.
Also, the dummy‑head’s response and each individual’s own response will obviously differ, and if there’s no crosstalk and the compensation is mismatched, you could end up hearing it as if it were suspended from the ceiling, as you described.
To listen to binaural through speakers (or HRIR/BRIR), you should ideally reduce or remove crosstalk—but since the ILD in far‑field recordings isn’t that extreme, you can just listen without too much trouble. Most binaural sources include DF compensation, so they’re rich in the ITD and ILD cues we care about, and that’s exactly what you’re hearing.
Of course, the YouTube link I shared might sound great, or it might sound bad, or it might evoke no sensation at all—it’s simply a difference. And yes, it feels closer to direct listening. If, based on that feeling, you recreate the space in a high‑quality, tightly controlled multichannel environment and record it on your body, then play it back through two speakers or headphones/IEMs, you can fully experience that spatial sense with only two channels. I think that’s a difference in type of recording, not in quality.
It is not as if I have one big and one small ear. ;-), but I do not have such nice castings of my ear canals. Cool stuff! My HRTFs look quite similar and in the room simulation I can use one BRIR for both ears with only moderate change.
Therefore my surprise with this drastic effect of the whole scene switching backwards (not left<->right as expected), but this was for an outdoor recording. Indoors the reflections provide a lot of room stability.
In the case of outdoor recordings, I can’t say exactly how that happens, but it’s an intriguing phenomenon. And yes—as you tested—you can use a single BRIR, adjust it for both ears, and still get a proper spatial image.
The human body naturally has asymmetries, and our brains learn from that data. Even so, if your left‑and‑right responses were to match perfectly, you’d still be able to hear directionality. (I’ve seen some people misunderstand this: they think the left‑ear response literally makes sounds come from the left. Of course, you need to account for reflections and their interactions in the recorded space, but occasionally people oversimplify and assume the IR itself “originates” on one side.)
)
